Hi.
On Fri, 29 Mar 2002, Tetsuo Koyama wrote:
> Dear FMRIB
>
> I would like to ask a question about group statistics.
>
> I ran a group comparison analysis between data obtained from different
> two tasks (Task A and Task B: these two were different in the scan time
> and numbers of stimuli to each other. Automatically, the predictive
> model functions also differed to each other). The first-level analyses
> had been already done separately on each scan basis. In FEAT group
> analysis, I assigned a number for the first-level output obtained from
> Task A (group A), and the rest number for first-level output obtained
> from Task B (group B). FEAT successfully yielded the result.
>
> In such a case with different predictive functions, what was exactly
> compared in the group statistics (signal change magnitude, fitting of
> the signal change to the predictive model, or other factors)? I need
> this information for interpretation of the result.
I need to make sure I understand your terms - are you saying that you had
a group of sessions/runs with task A and a different group with task B?
You are also saying that the EVs (explanatory variables / regressors /
columns in the design matrix) were different in the two cases as well?
So - when you do a two-group analysis, it is the parameter estimates
(which are always proportional to the change magnitude if the height of
the EV model is the same in all cases) that get compared across the
groups. If you use the fixed effects results then the first-level errors
get pooled to create the denominator in the resulting t-test. If you use
the random effects results it is the variance across sessions of the
parameters estimates that is used in the denominator.
This is all fine as long as the parameter estimates have the same
"meaning" in the two groups. Thus if two block-design models have the same
height in A and B but different timings, that is fine. An example of what
should NOT be done is if A was single-event and B was block - then the
parameters estimates cannot just be simply combined in the standard group
test.
I hope this makes sense - please feel free to ask for more detail if I
haven't answered your question.
Thanks, Steve.
Stephen M. Smith
Head of Image Analysis, FMRIB
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|